guide
iGaming retention CRM: the gradient boosting playbook (2026)
The strategic case for per-player scoring on raw transactions. Why cohort retention breaks, why not LLM, two-phase pilot scoping.
ReadWhat is broken
Four ways cohort retention leaks money in 2026.
Cohort retention leaks bonuses
Analyst-built segments push the same offer to everyone in the cohort. Bonus hunters get what VIPs should get. Margin walks out the door with the wrong audience.
VIP silence gets missed
A VIP whose bet size just dropped 60% is a churn signal that no cohort dashboard surfaces. By the time it shows up in tier movement, the window to hold them has closed.
Off-the-shelf CRMs do not fit
Your platform is self-written. Off-the-shelf retention CRMs assume a schema you do not have. Integration drags on for months, then still does not understand your player behaviour.
LLM scoring is not auditable
Vendors selling AI retention rarely explain how the scoring works. When the model flags a player, the analyst cannot see why. That kills adoption of the tool in the team that has to use it.
The stack we build
Six pieces that make retention actually work.
Per-player churn scoring
Gradient boosting on raw transactions. One decision per player, with the features that drove the flag. Analyst sees why, not just what.
Per-player reactivation scoring
For silent players, a second model returns reactivation probability if contacted. Outreach goes to players the model believes will respond.
Prioritised worklist into your CRM
Both scores flow into the retention team's existing CRM as a daily worklist. We do not replace the team. We upgrade the prioritisation.
Responsible gaming gate
Self-exclusion, deposit-limit, time-played flags read from your platform and checked before any retention trigger fires. No message reaches a player who has opted out or self-excluded.
Adapter for self-written platforms
We map your player schema to a standard feature card once. No migration off your existing stack. Integration in weeks, not months.
Optional outbound trigger
For operators without an internal retention team, the scoring service fires triggers into a connected dialer or messaging stack under your telco account.
Process
Two phases with an off-ramp.
01
Data audit and adapter (phase one)
We map your player schema, verify account-type flags, check consent state, validate time-window correctness. Two to four weeks. If the data is not ready to support the model, we tell you and stop rather than push to phase two.
02
Model and pilot (phase two)
Two scoring models trained on your six-plus months of history, offline evaluated on held-out windows, calibrated, integrated with your CRM in shadow mode for two weeks before analysts start acting on scores. Four to eight weeks.
03
Handover and retraining runbook
Full documentation of the feature pipeline, model artefacts, calibration procedure, and retraining cadence. You can operate independently. Optional monthly retainer covers drift monitoring and periodic retraining if you prefer.
// FAQ
Questions casino operators actually ask.
How is this different from off-the-shelf retention services?
Off-the-shelf services score cohorts and run generic robocalls scripted by the operator. We score per player, return the features that drove each score, and feed the output into the retention team's existing CRM as a prioritised worklist. The operator keeps the workflow and the analysts get the prioritisation.
Do you use LLMs for player scoring?
No. Scoring uses gradient boosting on raw transactions. Player payback runs five to seven months - a hallucinated risk score is real money. We use LLMs only where hallucination is recoverable: drafting retention copy for the team to review, summarising player history, generating human explanations of an auditable score.
Can you integrate with a self-written platform?
Yes. Almost every operator we work with runs a self-written platform. We build an adapter against your API or read your player database directly. Standard schema mapping: players, transactions, sessions, game events. No migration required.
How is responsible gaming handled?
Self-exclusion, deposit-limit, time-played, and other RG flags are read from your platform and checked before any retention trigger fires. Player data stays on your infrastructure. Specific regulatory frameworks - MGA, UKGC, regional licences - are scoped per deployment.
How much does a pilot cost?
Two-phase scoping. Phase one - data audit and adapter - is priced separately from phase two - model and pilot. We quote precisely after phase one, not before. If phase one shows the data is not ready to support the model, we stop and you pay only for phase one.
Bring the data. We do the audit before we quote.
First call is a thirty-minute scoping. We tell you honestly whether a pilot makes sense - and where custom development, or just a referral, fits better.
Book a 30-min call